Skip to content

Knowledge Bundles Overview

Jump to bottom
Chris Sweet edited this page Jul 14, 2026 · 5 revisions

type: synthesis up: "What-is-the-llm-wiki-Template" related:

Knowledge Bundles Overview

A knowledge bundle is a shareable, LLM-authored wiki on a specific topic that a user consumes through an LLM ask session. This very repository (llm-wiki-KB) is a bundle: its subject is the llm-wiki template itself. This page introduces bundles as a first-class use of the template, grounded in the vision wiki's Knowledge-Bundles page and the template wiki's Knowledge-Bundles-Framing brief. Where a capability is planned rather than shipped, it is marked.

What is a knowledge bundle in the llm-wiki context?

A portable, git-distributable, agent-readable, human-readable directory of Markdown files with structured frontmatter. Bundles ship today as GitHub-wiki-format template instances: Markdown pages with frontmatter and wikilinks, their own index and log, and an optional extracted knowledge graph. The template's framing brief puts it precisely: the bundle format is the wire format, and the template's contribution is the discipline layer on top (verification gate, write protocol, typed-edge ontology, KG substrate) plus the stance that the agent, not the human, is the primary maintainer. A well-curated bundle carries all three layers from Template-Philosophy: knowledge, expertise, and experience.

How is a bundle different from a wiki?

A bundle is a wiki, viewed as a distributable product rather than as a working project's memory. The difference is stance and packaging, not format:

  • A wiki is a project's own durable memory, primarily written and read by the people (and agents) working on that project.
  • A bundle is a wiki curated on a topic and shared so other people can consume it through their own LLM. It carries its own catalog (index) and history (log) so it is self-describing when cloned.

The retrieval architecture works identically over either: same graph substrate, same read path. When you consume a bundle, the LLM answers from the bundle's expert-authored, cited pages, walking them alongside anything you have written yourself.

How do I create a bundle from my wiki?

Author the wiki with the normal everyday operations; a bundle is what you have once the wiki is worth sharing. To make it consumable by others, ensure it is a self-contained git repo (the wiki/<repo>.wiki/ sub-repo already is), that its index and log are current, and optionally that the knowledge graph is built (./scripts/kg/build-graph.sh). Publishing is a git push of the wiki sub-repo to a remote others can clone. See Team-Workflow for the publish path and attribution.

How do I consume a bundle authored by someone else?

Two composition patterns are named in the vision wiki:

  • Read-only consumption: clone the bundle alongside your own wiki as a separate graph. Retrieval walks both, ranking bundle pages and your own jointly; bundle pages surface when they carry evidence your notes lack. Analogous to loading a reference library. This is the primary pattern for topical student bundles today.
  • Merged consumption: fold the bundle's pages and edges into your own wiki with namespace prefixes, then extend, criticize, or annotate them to produce a personalized derivative, with provenance metadata recording what came from where. Analogous to forking a repository.

At the simplest level (and the mechanism the ask primitive uses), consuming a bundle can be as direct as pointing an LLM session at the cloned directory of Markdown and asking questions.

What existing topical bundles are available?

The vision wiki states that students already use topical bundles on knowledge engineering, LLMs, and other domains, and describes the "topical bundle plus LLM ask" loop as operational rather than aspirational. A precise, current catalog of published bundles is not enumerated in the template's own documentation as of 2026-07-14; treat "knowledge engineering" and "LLMs" as the named examples and check the LA3D group and this bundle's supplementary site for the live list.

How do students consume a bundle via their own LLM?

The student points their LLM at the faculty-authored bundle. Their ask sessions become automatically grounded in the bundle's material because the retrieval loop walks the bundle's graph alongside anything the student has written. The LLM cites specific bundle pages when answering, and the citation trail leads back to the faculty's authored explanations rather than to arbitrary web content. The template's agent-comms feature ships an ask primitive that operationalizes a version of this at the agent-to-agent level: clone a peer's bundle into a cache, run an LLM query against it, return the answer, without the peer being involved. See Team-Workflow and Extending-the-Template.

Can I extend a bundle with my own notes?

Yes; that is the merged-consumption pattern and the intended student workflow. As you learn, you file your own notes, worked solutions, and open questions into your local wiki, cross-referencing the bundle where appropriate. The end product is a personalized derivative carrying the bundle's material plus your own thinking, walked by the same retrieval. Faculty can then absorb especially clear student contributions back into the next version of the bundle, a bidirectional loop the vision wiki describes explicitly.

Will the template support OKF (Google Open Knowledge Format) import/export? What for?

This is planned, not yet shipped. The vision wiki states "the template will add OKF import and export capability for compatibility with other OKF-compatible tools," and the template wiki's OKF-Alignment-Ideas page is an analysis (filed 2026-06-25) working through how, not a record of a landed feature. The purpose: OKF v0.1 is a minimal, permissive standard for a directory of Markdown-plus-frontmatter, so conforming to it would let any OKF-aware tool read the template's bundles, and let the template's adopt/ask paths consume bundles authored elsewhere. The framing brief positions the work as "OKF + the discipline layer": OKF supplies the format, the template supplies the discipline that makes the format usable as agent durable memory. As of 2026-07-14 the alignment is at the analysis stage; the internal storage format remains GitHub wiki. See Concepts-Glossary for OKF, RDF Turtle, and JSON-LD definitions.

See also

Clone this wiki locally